Taiwan Stock One-Click Financial Model & Investment Deck Generator
Enter any Taiwan stock ticker and it instantly gathers public filings and daily price data to build an auditable Excel model plus a full investment presentation.
Evidence shows no filesystem overreach, no credential collection, no covert exfiltration; source attribution, disclaimers and a Sources sheet are required — positive. Deductions: the mandate to always execute the full flow ('must not reply with text only') weakens user control; data comes from unverified web_search; no least-privilege declaration or confirmation mechanism; license metadata unknown.
The workflow is self-consistent with recalc.py error checks, spot-check rules and a pitfalls section. Deductions: hard dependency on /mnt/skills/public/*, web_search and present_files means it cannot run outside the Claude container; failure handling relies on 'estimate and label'; no tests or runnable script samples committed.
Triggers, inputs and outputs are clearly specified, Traditional Chinese friendly, audience is concrete. Deductions: the trigger list is overly broad (generic terms like '估值模型' risk false triggers); environment is bound to Claude container paths; mainland-China reachability of TWSE/search backends unevaluated; non-fit ranges (US stocks etc.) undeclared.
Docs are well layered (workflow, specs, pitfalls, install guide, disclaimer). Deductions: license metadata missing; no version number, changelog or maintainer commitment; INSTALL details referenced but outside assessed files; external skill docs depended on but not included.
Output specs (8-sheet Excel, 14-slide PPT) are detailed with a clear non-advisory positioning. Deductions: static review cannot verify output quality; 'zero formula errors' is an instruction, not evidence; quality depends heavily on the host model's search capability; no sample outputs committed.
Mandated source labeling and recalc checks are auditable by design. Deductions: no committed test suite, no sample artifacts, no third-party execution evidence; all claims remain at instruction level and cannot be reproduced statically.
- Hardcoded dependency on /mnt/skills/public/ and the container environment means the skill will likely not run as described elsewhere.
- No license metadata, versioning or maintenance info; verify repository license and update status before use.
- Financial data comes from live web search with no accuracy guarantee; verify primary sources before any investment decision.
- Overly broad trigger list may auto-invoke the full time-consuming flow unexpectedly.
- Core dependencies (TWSE, search engines) are overseas services; mainland-China reachability is unverified.
What does this skill do, and when should you use it?
This is an Agent Skill for Claude / Codex that turns a Taiwan stock ticker into complete analysis deliverables. It searches live public financial data and at least 60 trading days of daily prices, then produces a formula-linked Excel financial model (three-scenario forecasts, P/B and P/S valuation, sensitivity tables, KD/MACD indicators, risk matrix) and a 14+ page investment PPTX, with downloadable candlestick-pattern HTML/PNG charts when generated. Its stated positioning is analysis and teaching aid, not automated investment advice — every delivery must include honest findings and a disclaimer. It suits anyone who needs Taiwan stock public data turned into adjustable, presentable material quickly.
On trigger it runs a five-step pipeline: (1) performs 3-5 web searches to collect three years of revenue/EPS, five quarters of EPS, monthly revenue, BVPS, latest price and 3-7 risks, with source and date noted for every hardcoded number; (2) pulls daily bars from official TWSE/TPEx data and computes MA5/10/20, Bollinger Bands (20,2), KD (9,3,3), MACD, and rule-based candlestick pattern markers (breakouts, gaps, doji, golden/death crosses); (3) builds an Excel workbook of at least 8 sheets with openpyxl, computing everything via Excel formulas (Python-side hardcoding is banned), then runs recalc.py requiring total_errors: 0; (4) builds a 14+ page PPTX with pptxgenjs in a Midnight Executive palette and Microsoft JhengHei, using native addChart only, followed by rezip.py and per-slide visual QA; (5) delivers the .xlsx model, .pptx deck (and .html/.png candlestick charts if produced) to the outputs directory with honest findings and a 'not investment advice' note.
- A retail investor who wants a quick valuation and three-scenario price targets for a Taiwan stock like 2330 or 6116 without manually compiling filings
- A finance instructor or study group needing an adjustable, formula-traceable Excel model as teaching material
- An analyst adding KD/MACD and candlestick-pattern slides to cover short-term technical positioning
- Covering loss-making or cyclical (panel/DRAM) stocks, where a P/B-centric valuation framework and wider assumption ranges are appropriate
- Compiling monthly revenue, quarterly EPS, a risk matrix and cited sources into a single presentable deliverable
What are this skill's strengths and limitations?
- Complete output structure: at least 8 Excel sheets and 14 slides covering financials, valuation, technicals and risk
- Strong auditability: every hardcoded number carries source and date, all math uses Excel formulas, and recalc.py must report zero errors
- Respects Taiwan chart conventions (red = up, green = down) with explicit hard color rules
- Data-integrity design: derived values must state their method, rule-based pattern detection must be labeled, deliveries must include honest findings and a disclaimer
- Depends on web_search and TWSE/TPEx data availability; on conflicting search results it defers to the newest date, so data quality hinges on live search
- References Claude-platform built-in script paths (/mnt/skills/public/...), which other or self-hosted environments must replicate
- License is unknown — verify before commercial use
- No test suite or sample outputs to verify quality; candlestick patterns are rule-based detection, explicitly not trading signals
- Not usable directly from a bare model API without a shell plus Python/Node toolchain
How do you install this skill?
Codex: copy skills/ai-stock-report-generator/SKILL.md to ~/.codex/skills/ai-stock-report-generator/SKILL.md (the README includes Windows PowerShell New-Item / Copy-Item examples), then restart Codex. Claude Code: place it at .claude/skills/ai-stock-report-generator/SKILL.md in your project, or ~/.claude/skills/ai-stock-report-generator/SKILL.md. Full instructions are in docs/INSTALL.md. Note the SKILL.md references /mnt/skills/public/xlsx and pptx helper scripts (recalc.py, rezip.py) — these are Claude-platform built-in paths; other environments must supply equivalents.
How do you use this skill?
Provide a Taiwan ticker (4-6 digits) with a trigger, e.g.: 2303 //ai-stock-report-generator; or natural language like "build a valuation and deck for 2303" or "make a financial model for 6116 as a PPT". The skill runs the full pipeline and outputs {ticker}_{name} financial model .xlsx and investment analysis .pptx (plus candlestick chart .html/.png if generated). Editing the blue input cells in the Excel 'Assumptions' sheet re-drives the entire workbook.